Skip to content

vllm_omni.diffusion.model_loader.gguf_adapters.z_image ¶

Z_IMAGE_KEYS_RENAME_DICT `module-attribute` ¶

Z_IMAGE_KEYS_RENAME_DICT = {
    "final_layer.": "all_final_layer.2-1.",
    "x_embedder.": "all_x_embedder.2-1.",
    ".attention.qkv": ".attention.to_qkv",
    ".attention.k_norm": ".attention.norm_k",
    ".attention.q_norm": ".attention.norm_q",
    ".attention.out": ".attention.to_out.0",
    "model.diffusion_model.": "",
}

ZImageGGUFAdapter ¶

Bases: GGUFAdapter

GGUF adapter for Z-Image models with QKV/FFN shard support.

gguf_to_hf_mapper `class-attribute` `instance-attribute` ¶

gguf_to_hf_mapper = WeightsMapper(
    orig_to_new_substr=Z_IMAGE_KEYS_RENAME_DICT
)

is_compatible `staticmethod` ¶

is_compatible(od_config, model: Module, source) -> bool

weights_iterator ¶

weights_iterator() -> Iterable[tuple[str, Tensor]]